TSMC 5nm fewer defects than 7nm at equivalent time

[porina]

 

Quote

the current N5 process from TSMC has a lower defect density than N7 did at the same time in its development cycle. TSMC. This slide from TSMC was showcased near the start of the event

Source

https://www.anandtech.com/show/16028/better-yield-on-5nm-than-7nm-tsmc-update-on-defect-rates-for-n5

 

Summary

TSMC are indicating that the defect rate of their 5nm process is doing better than 7nm was at a comparable time in its life cycle relative to the introduction to High Volume Manufacturing. This is part attributed to the move to EUV, which reduces complexity in the process compared to the multiple steps of DUV required previously.

 

Thoughts

TSMC are the current leaders in silicon device production and this should help keep them in that spot, and also benefit those who use them to manufacture their chips. A node advancement brings with it advantages, some of which are also shown in the slide. It will be really interesting to see what hardware we get once that hits PC tech mainstream in the not too distant future. It is also interesting to see 6nm on that slide, which Intel have ordered some of for future production.

 

 

[BobVonBob]

Very interesting, although the gap between N7 and N5 seems to be closing, with N5 looking like it will level off higher than N7. On the other hand they only have three data points and I'm probably extrapolating too much from them.

[porina]

2 minutes ago, BobVonBob said:

Very interesting, although the gap between N7 and N5 seems to be closing, with N5 looking like it will level off higher than N7. On the other hand they only have three data points and I'm probably extrapolating too much from them.

The measure is defects per silicon area. At the same defect rate, the yields should go up relative to 7nm as the 5nm device will be a smaller area. TSMC also have to balance costs. If it'll make the process more expensive to reduce the defect rate, it might not be a great tradeoff. Ongoing improvements are expected to a point though. It is still relatively new.

[Drama Lama]

Come on Intel 

 

TSMC has also N3 ready for 2022

Edited August 25, 2020 by Drama Lama

[Vishera]

Intel and Global Foundries are still stuck at 14nm...

[GDRRiley]

11 minutes ago, Vishera said:

Intel and Global Foundries are still stuck at 14nm...

Global foundries gave up on 7nm for now.

Intels got 10nm working somewhat. they have some products on it

[Vishera]

4 minutes ago, GDRRiley said:

Intels got 10nm working somewhat. they have some products on it

Only mobile devices... They are having trouble in producing 10nm in the quantities they need so it's only for mobile.

[soldier_ph]

3 hours ago, Drama Lama said:

Come on Intel 

 

TSMC has also N3 ready for 2022

At this point they can't see Intel in the rear view Mirror anymore.

[Arika]

Good to see TSMC pushing so hard into this new tech is always good, but i can't help but feel it will turn around to bite them hard. they have a LOT of big name clients who are going to want to jump on this asap, they will not be able to keep up with demand now that they are also responsible for taking on the Apple chips as well. NVidia already had to go to Samsung for some of their stuff.

[bcredeur97]

6 hours ago, Vishera said:

Only mobile devices... They are having trouble in producing 10nm in the quantities they need so it's only for mobile.

iirc intel kinda self-enforces higher standards on their processes than TSMC does, correct? Like some of the other clearances on things are tighter. 

If this is the case, I do think TSMC has the better idea tho. Just get the process node down as much as possible now, work on perfecting it once you're completely stuck with physical limitations afterwards.

[Drama Lama]

4 hours ago, Arika S said:

NVidia already had to go to Samsung for some of their stuff.

But another reason why Nvidia went for Samsung is because 8nm Samsung  is cheaper than 7nm TSMC and they want to keep their high margins 

[Liltrekkie]

6 hours ago, Drama Lama said:

But another reason why Nvidia went for Samsung is because 8nm Samsung  is cheaper than 7nm TSMC and they want to keep their high margins 

Nvidia charges what they want to for a GPU, they could switch to TSMC and bump the price up accordingly, no idea why they won't, Nvidia is the only choice for high end gaming, above 2070 super anyway, which the 5700xt is neck and neck with. 

 

It seems odd, because there was a rumor back in may that Nvidia secured orders for 7nm and 5nm TSMC chips. So I wonder which is true, or if they are both true, which products will be done on which node.

 

Now if only AMD could get it's head out of the sand and design a decent GPU, maybe this outrageous pricing crap would be over with

[porina]

1 hour ago, Liltrekkie said:

It seems odd, because there was a rumor back in may that Nvidia secured orders for 7nm and 5nm TSMC chips. So I wonder which is true, or if they are both true, which products will be done on which node.

It might not be yes/no, but a case of how much capacity? 5nm is still a (near) future technology so wont be applying immediately to next gen products. 7nm we already knew was a great seller for TSMC so it doesn't sound like they're short on customers there. For whatever (multiple) reasons nvidia may not have got access to as much as they can use.

 

1 hour ago, Liltrekkie said:

Now if only AMD could get it's head out of the sand and design a decent GPU, maybe this outrageous pricing crap would be over with

Lots of hype in next gen consoles, so next Navi iteration should be interesting to keep an eye on. I wouldn't hold my breath for an nvidia beating top card, but if as expected they start to catch up on feature and improved efficiency, they should still be potent options in the mainsteam.

[Vishera]

11 hours ago, bcredeur97 said:

iirc intel kinda self-enforces higher standards on their processes than TSMC does, correct? Like some of the other clearances on things are tighter. 

True.

11 hours ago, bcredeur97 said:

If this is the case, I do think TSMC has the better idea tho. Just get the process node down as much as possible now, work on perfecting it once you're completely stuck with physical limitations afterwards.

I disagree,Intel's new 10nm SuperFin technology is superior to TSMCs 7nm FinFET.

The problem is that Intel doesn't convert 14nm FinFET fabs to 10nm SuperFin.

[Doobeedoo]

Pretty great. 

[porina]

39 minutes ago, Vishera said:

The problem is that Intel doesn't convert 14nm FinFET fabs to 10nm SuperFin.

Upgrades are neither cheap nor fast. Intel 10nm is late for volume production now. If they aren't already well down that road with increased 10nm capacity, they have to plan ahead for 7nm and beyond.

 

We still have to wait and see what real world performance Tiger Lake brings, as the first mass product based on the updated 10nm. But in a year or two we're into 7nm territory anyway.

[StDragon]

The problem is that TSMC and Intel define the parameters of x-nm sizes differently. So it's not a good comparison to begin with. That said however, there's no doubt that Intel is behind the curve with regards to next generation fabrication.

[Drama Lama]

2 hours ago, Vishera said:

disagree,Intel's new 10nm SuperFin technology is superior to TSMCs 7nm FinFET.

The problem is that Intel doesn't convert 14nm FinFET fabs to 10nm SuperFin.

and they had too many defects 

and by now it was planned that most of intel‘s processors would be on 10nm so they didn’t order new 14 nm machines because if everything went like planned they wouldn’t need them for high end products ( correct me if I’m wrong)

[Spindel]

Now just release those sweet sweet A-series AS chips!

[uzzi38]

On 8/25/2020 at 7:20 PM, gabrielcarvfer said:

Sweet sweet EUV. Now waiting for 3D stacking.

Thankfully we're very, very close already.